Waar begin je?
Begin met je hardware of gewenste modellen. Vergelijk capaciteit, prestaties en bewijs vóór installatie.
Select hardware to see model fit results.
Guides & methodology
Browse the complete catalogs and estimation method when you need them.
Alle hardware21
Elke klasse heeft een eigen budget, runtimepad en realistische bovengrens.
Mac Studio M5 Ultra · 96 GB
The entry M5 Ultra configuration combines a 96 GB unified pool with 1.2 TB/s memory bandwidth. Apple announced it in August 2026; availability begins September 22.
Open gidsMac mini M6 · 24 GB
A compact current-generation Mac with 24 GB unified memory. Keep macOS, runtime, and KV-cache headroom separate from model weights.
Open gidsMacBook Pro M5 Max · 64 GB
A portable Apple-native tier for larger quantized models. Sustained inference still depends on thermals, runtime support, and context length.
Open gidsFramework Desktop · Ryzen AI Max+ 395 · 128 GB
A 128 GB Strix Halo desktop. AMD allows a large graphics-memory allocation, but usable capacity and speed depend on BIOS, driver, and runtime support.
Open gidsRadeon AI PRO R9700 · 32 GB
A 32 GB AMD workstation card for larger single-GPU quantized models. Check the exact operating-system and ROCm or Vulkan runtime path first.
Open gidsNVIDIA DGX Spark · 128 GB
A compact local AI system with 128 GB coherent memory. It can hold large low-bit models, but decode speed and software support remain model-specific.
Open gidsMac Studio M3 Ultra · 96 GB
A quiet single-box option for 70B-class Q4 models and Apple-native MLX workflows, with memory shared by macOS and applications.
Open gidsMac Studio M3 Ultra · 256 GB
Enough capacity for many 100B-plus low-bit models, provided the architecture is supported and optimized by the selected runtime.
Open gidsGeForce RTX 5090 · 32 GB
A fast single-GPU tier for 27B-class low-bit models, with limited remaining space for very long context or multiple concurrent users.
Open gidsGeForce RTX 4090 · 24 GB
A 27B Q4 checkpoint can be a tight fit. Leave room for runtime overhead and KV cache instead of filling every gigabyte with weights.
Open gids8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
Open gids16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
Open gidsApple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
Open gidsNVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
Open gidsNVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
Open gidsNVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
Open gidsApple silicon · 32 GB
24B–32B Q4 models become practical for one local user.
Open gidsNVIDIA · 24 GB
The consumer sweet spot for 27B and 30B-A3B INT4/Q4 models.
Open gidsApple silicon · 64 GB
32B models with long context or 70B Q4 with conservative headroom.
Open gidsWorkstation · 48 GB
70B Q4 serving begins here, with limited room for cache and concurrency.
Open gidsDatacenter · 80 GB
Large MoE models and high-throughput servers, still bounded by quantization and cache.
Open gidsAlle modellen24
Qwen3.8 Flash Next
English- Geschat geheugen
- ~108 GB
- Context
- 262,144 native; 1M extended
Ling-3.0-flash
English- Geschat geheugen
- ~74.4 GB
- Context
- Check selected runtime and checkpoint
Nex-N2.5-mini
English- Geschat geheugen
- ~21 GB
- Context
- 262,144 (config max_position_embeddings)
Qwen3.6 27B
- Geschat geheugen
- ~18.5 GB
- Context
- 262K native
Qwen3.8 27B
- Geschat geheugen
- ~19.5 GB
- Context
- 262K native
Qwen3.8 Flash Next REAP-288
- Geschat geheugen
- ~40.57 GB
- Context
- 7.6K recommended on measured 48 GB Mac
Qwen3.6 35B-A3B
- Geschat geheugen
- ~25 GB
- Context
- 262K native
DeepSeek V4 Flash
- Geschat geheugen
- ~176 GB
- Context
- 1M native
Gemma 3 1B
- Geschat geheugen
- ~1.4 GB
- Context
- 32K
Llama 3.2 3B
- Geschat geheugen
- ~2.8 GB
- Context
- 128K
Qwen3 4B
- Geschat geheugen
- ~3.6 GB
- Context
- 32K+
Gemma 3 4B
- Geschat geheugen
- ~4.2 GB
- Context
- 128K
Qwen3 8B
- Geschat geheugen
- ~6.8 GB
- Context
- 32K+
Gemma 3 12B
- Geschat geheugen
- ~9.4 GB
- Context
- 128K
Qwen3 14B
- Geschat geheugen
- ~11.2 GB
- Context
- 32K+
gpt-oss-20b
- Geschat geheugen
- ~16 GB
- Context
- 128K
Devstral Small 2 24B
- Geschat geheugen
- ~18.5 GB
- Context
- 256K
Gemma 3 27B
- Geschat geheugen
- ~20.5 GB
- Context
- 128K
Qwen3 30B-A3B
- Geschat geheugen
- ~22.5 GB
- Context
- 32K+
Qwen3 32B
- Geschat geheugen
- ~24.5 GB
- Context
- 32K+
Llama 3.1 70B
- Geschat geheugen
- ~46 GB
- Context
- 128K
Llama 4 Scout
- Geschat geheugen
- ~78 GB
- Context
- 10M advertised
gpt-oss-120b
- Geschat geheugen
- ~80 GB
- Context
- 128K
Qwen-Image 2.1
English- Geschat geheugen
- ~33.1 GB
- Context
- Not a token context: one prompt plus up to 10 reference images
Kies de inferentielaag15
Ollama
One-command local chat and app integration
Open gidsllama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
Open gidsLM Studio
Discovering, downloading and testing models without a terminal
Open gidsMLX LM
Native Apple silicon inference, experimentation and fine-tuning
Open gidsoMLX
Apple-silicon serving with continuous batching, memory guards, and supported SSD-backed paths
Open gidsvLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
Open gidsTensorRT-LLM
Maximum NVIDIA throughput after engine tuning
Open gidsSGLang
Multi-request GPU serving with prefix caching and speculative decoding
Open gidsLocalAI
One OpenAI-compatible endpoint for text, vision, speech and image models, GPU optional
Open gidsLemonade
Ryzen AI, Radeon and Strix Halo PCs, including the XDNA2 NPU, behind one local API
Open gidsOpenVINO Model Server
Serving OpenVINO IR models on Intel CPUs, Arc GPUs and NPUs with one API
Open gidsKoboldCpp
One-file GGUF chat on Windows, macOS or Linux with no Python install
Open gidsllamafile
Distributing a whole GGUF model as one runnable executable on Linux, macOS, Windows and BSD
Open gidsKTransformers
Very large MoE models with hot experts on the GPU and cold experts in system RAM
Open gidsUnsloth
Running GGUF or MLX models from a desktop app that also trains and exports them
Open gidsOnze schatting
We gebruiken gedocumenteerde groottes en quantisatie, reserveren ruimte voor systeem en runtime en tellen cache extra. MoE gebruikt totaalparameters.
We scheiden totale parameters, actieve MoE-parameters, quantized fit, ondersteuning en gemeten prestaties.
- Memory fit is an estimate, not a speed measurement.
- Size MoE by total parameters; shared and distributed memory need additional checks.
- Vendor claims, community measurements, runtime conditions and unknowns are disclosed separately.
Catalog data updated: 2026-10-03
Voor je 40GB downloadt
Does a 24 GB GPU run a 30B model?
Often at Q4/INT4 with a conservative context. Qwen3 30B-A3B and Gemma 3 27B are representative fits, but cache and runtime overhead still matter.
Are active MoE parameters the memory requirement?
No. Active parameters affect compute per token; total parameters still need to be stored in memory or offloaded.
Which runtime should a beginner choose?
Ollama for a terminal-first setup or LM Studio for a visual desktop. llama.cpp is the portable fallback; vLLM is for higher-throughput GPU serving.
How do you report speed?
Exact device-and-model measurements are labeled measured. Relative manufacturer figures are labeled vendor claims. Every other combination stays unknown instead of receiving an invented tokens-per-second estimate.